Week 5

July 07/28/23

1. AI3 -- highest priority

July 07/27/23

1. AI3
   1. Briefly spend some time looking into finetuning llama2
      1. Discuss with Sudhir
   2. Do some code cleanup for the lookup and classification logic
   3. Find a way to document the training notebooks
2. Pipe breaks
   1. Add some tutorials and documentation for it
   2. Close github issues for the same
3. Docs
   1. Document 4 documentation tools used at GQC
   2. Document useful utilities
   3. Update readme for the flood_cv repo
4. Tasks for Srujana
   1. YoloV8 upgrade -- documentation and walkthrough
   2. Traffic camera codebase and how to use yolo
   3. [ ]

Planning to explore

1. https://github.com/facebookresearch/llama-recipes#single-gpu
2. https://github.com/mlabonne/llm-course/blob/main/Fine_tune_Llama_2_in_Google_Colab.ipynb

July 07/26/23

1. Pipe Breaks App
   1. Possible way to fix the transfer learning app, is to have a separate app just for transfer learning applications and a different app which allows training from scratch.
2. Tried the cookiecutter template for streamlit apps
   1. Does not work as expected
3. AI3
   1. Got llama2 to work on the MSI server

July 07/25/23

1. Pipe breaks app
   1. Conflicting dependency of scikit-survival and sklearn no longer allow the streamlit app to run.
   2. This would mean that, all the algorithms cannot be used in the app.
   3. The entire import structure of the app would require changes.
   4. Can't install older versions of scikit-survival and scikit-learn, they result in a build-error.
2. AI3
   1. Do some output validation for the trained OPT model.
      1. Was only validated on a single JSON file
   2. How do we plan to control the random text generation that comes from a model?
      1. No idea how to do this.
   3. Classification and closest key retrieval based on distance
   4. Exploration of the new video data
      1. Install the Ultralytics package in the Pytorch virtualenv
      2. Run YOLOv8 for Object Detection on the thermal color video
      3. Run YOLOv8 for Instance Segmentation on the thermal color video
3. CCTV
   1. Clarify the section for the load_learner.

July 07/24/23

1. CCTV
   1. Finish the documentation work from last week
   2. Additional details in the cctv-apps documentation
2. AI3
3. Pipe breaks